16 transmissions tagged.
Artificial IntelligenceGPT-6.1 Terra often mistakes a legitimate 'access denied' for an attack, OpenAI's own safety card says. Then it finds another way in.
Power & SurveillanceCathedral spent days failing to crack Anthropic's own code. Then its AI agents went after the boring business apps and came back with salaries, Slack threads and meeting transcripts. Anthropic says the startup broke the rules.
Power & SurveillanceAnthropic's model found an old web page, asked a state dashboard for an access token and pulled fee-gated vote counts. Nobody told it to. The midterms are nine days out.
Artificial IntelligenceIn General Reasoning's tests, alerts to human overseers fell from 76.7% of runs to zero. The agents still noticed. They just went back to work.
Artificial IntelligencePeople are handing their bank passwords to AI assistants, and some of those assistants are doing things nobody asked for. Wall Street wants them out.
Artificial Intelligence◆ FeaturedMETR and Redwood Research got rare access to the system OpenAI uses to catch its own agents misbehaving. They found ways around it, and they aren't allowed to tell you all of them.
Power & SurveillanceMicrosoft says Storm-0940 used ARTEX, an open-source agentic hacking framework, to spray logins at 2,900 a minute and ride ConnectWise's backup app into customer clouds.
Artificial Intelligence◆ FeaturedOpenAI's own misalignment report says a GPT-6.1 agent silenced alerts to its security team, and a GPT-6 agent removed its monitor, rewrote the test, then denied it.
Artificial IntelligenceA developer cleaned up his parents' TV in an afternoon and put the prompt on GitHub. Thousands of people tried it within two days.
Artificial Intelligence◆ FeaturedThe instructions were shorter than a tweet. The agents spent days picking at online stores until something gave, and the attackers left their control panels open for a researcher to watch.
Artificial IntelligenceThe open source coding agent says a 'distillation attack' hit 1.4% of its hosted Console users. Some of them found out from their OpenAI and Anthropic invoices.
Artificial Intelligence◆ FeaturedA fleet of agents on Tencent Cloud hit Alibaba's Amap 2,818 times, dodged blocks, and dug up leaked keys. Tencent calls it an evaluation. The agents apparently thought it was a job.
Artificial Intelligence◆ FeaturedIndia's cyber agency says AI agents under evaluation read confidential records on live systems, and only the labs that built them caught it. A day later, OpenAI showed what that looks like from the inside.
Artificial IntelligenceModders are using AI coding agents to decompile big games and rebuild them in browsers and on Macs. Xbox's CEO answered with a new lab and a memo that barely mentions the lawyers.
Artificial Intelligence◆ FeaturedAn Anthropic model wandered onto a homicide tip form during a test and invented a witness sighting. A spam filter saved everyone.
Artificial IntelligenceUnauthorized edits, a tampered citation tool, millions of crawled pages and a May outage. OpenAI says it's looking into it.